Commits

Show all
Author Commit Message Labels Comments Date
Daniel Roberts
We now make it through rtyping, and rlib/test/test_jit.py passes.
Branches
ootype-virtualrefs
Daniel Roberts
Passing all tests, changed ooupcast to oodowncast and fixed a test helper function.
Branches
ootype-virtualrefs
Daniel Roberts
Passing another test by implementing OOVRefRepr.convert_const(), hopefully correctly.
Branches
ootype-virtualrefs
Daniel Roberts
Properly test interp() results.
Branches
ootype-virtualrefs
Daniel Roberts
Fixed the tests a bit more.
Branches
ootype-virtualrefs
Daniel Roberts
Made tests run on both lltypesystem and ootypesystem.
Branches
ootype-virtualrefs
Daniel Roberts
SomeVRef now has a more correct Repr on ootypesystem
Branches
ootype-virtualrefs
Daniel Roberts
Hopefully this branch will lead to fixing ootype translation.
Branches
ootype-virtualrefs
Amaury Forgeot d'Arc
Attempt to define a "scoped allocator" to use in a 'with' statement.
Antonio Cuni
(antocuni, arigo around): finally fix test_unsigned_short_arguments on 64 bits, by handling all the possible cases separately.
Antonio Cuni
(antocuni, arigo): fix the test: it relied on the fact that dummy was 0 at the beginning of the test, but this is not necessarly true if the same shared library is used by someone else (like, metainterp/test/test_ffi_call)
Amaury Forgeot d'Arc
Improve comment
Amaury Forgeot d'Arc
Move SETUP_WITH out of the Unimplemented section
Amaury Forgeot d'Arc
Implement the SETUP_WITH opcode for the flow object space running on code compiled by cpython 2.7. Now all previous tests run on top of cpython 2.5 to 2.7. The implementation was borrowed from the fast-forward branch, and slightly simplified to adapt to the RPython type system, which has virtual methods but no __mro__ or special descriptors. This is why this code is not in pyopcode.py
Amaury Forgeot d'Arc
Add support for "with" statements in RPython. The goal is to simplify resource management, like temporary allocations, scoped locks, etc:: with lltype.alloc_buffer(30) as tmpbuf: [...use tmpbuf...] The __exit__ function is always called, even when an exception is raised. Actually there was not much to do (a space.wrap() was missing) ... at least on top of CPython2.5 It's more complex for other versions of CPython, because this opcode change implementations every release...
Antonio Cuni
make rlib.libffi more robust when calling a function that it's declared to return a smaller unsigned, but then called expecting a ULONG. This fixes test_unsigned_short_args in module/_ffi
Armin Rigo
Fix the test on 64 bits.
Armin Rigo
Reorder a bit the code in item_w(). It should help the JIT generate less operations, because it doesn't need to compare the item with the lower and upper bounds explicitly. It only needs to do the casting (which requires masking) and then comparing the result with the original value.
Antonio Cuni
checkin this test which passes on 32bit, but I expect it to fail on 64
Armin Rigo
Use $CC on mac too. Fixes test_gcc_options.
Armin Rigo
Fix test.
Armin Rigo
Fix the tests, at least on Mac.
Armin Rigo
- add a missing "const" - add the symbol "ret_un_func" to the list of exported symbols (needed on Mac OS/X, for example)
Armin Rigo
Test and fix.
Armin Rigo
A non-rpython version of str2charp that allocates an immortal string.
Armin Rigo
Fix order-dependency annotation issues.
Armin Rigo
Fix for 64-bits.
Armin Rigo
Fix.
Armin Rigo
(antocuni, arigo) Hopefully fix lib-python/.../test_descr on 64-bit platforms. The issue is that we can get a cycle of tail recursions (more easily on x86-64 than on x86-32), and in such cycles, the stack overflow detection does not work as a way to interrupt the cycle. Thus we can get caught in what seems like an infinite recursion (but is not in C).
Armin Rigo
'track_allocation=False' must also be given on lltype.free().
  1. Prev
  2. Next